[apologies for cross-posting] Terminology Translation Task @WMT2026 �C Call for Participation We are excited to announce the 4th Terminology Translation Shared Task, which will be held as part of the 11th Conference on Machine Translation (WMT 2026), co-located with EMNLP 2026 in Budapest, Hungary. TL;DR
* Two tracks on terminology-aware document-level machine translation. * Language pairs: Spanish��Basque, English��Polish, and Traditional Chinese��English across engineering, medicine, and finance domains. * Evaluation metrics: both general MT quality and terminology-oriented metrics. * Task started: 30 June 2026 * System submission deadline: 31 July 2026 (AoE) * Shared task website: https://www2.statmt.org/wmt26/terminology.html * Pre-registration: https://forms.gle/nCHmek9UkdJaraqu9
OVERVIEW Despite remarkable progress in machine translation, accurately translating specialized terminology while maintaining overall translation quality remains a challenging problem. Previous editions of the Terminology Translation Shared Task demonstrated that document-level terminology translation is still substantially more difficult than sentence-level translation, particularly when consistent terminology usage is required throughout a document. This year's edition introduces two major developments. First, we expand the shared task to lower-resource, morphologically rich languages: Basque and Polish. Second, alongside our established terminology-dictionary setting, we introduce a new track in which systems must infer terminology from seed bitexts, providing a more realistic and challenging evaluation scenario. We welcome submissions from both academia and industry and impose no restrictions on models or training resources.
IMPORTANT DATES (AoE)
* 1 Jul 2026: Test data and submission form released; shared task begins * 31 Jul 2026: System submission deadline * 7 Aug 2026: paper description submission (according to the WMT 2026 schedule) * 28-29 October 2026: WMT 2026 conference
SUBMISSION TRACKS Track 1: Document-Level Translation with Explicit Dictionary Participants receive document-level source texts together with terminology dictionaries and evaluate their systems under different terminology conditions. Track 2: Document-Level Translation with Seed Bitexts Instead of explicit terminology dictionaries, participants receive terminology-rich parallel examples ("seed bitexts") and must leverage them to translate related documents.
LANGUAGES AND DOMAINS
* Spanish �� Basque (Engineering & Technology), * English �� Polish (Engineering & Technology, Medicine), * Traditional Chinese �� English (Finance)
EVALUATION Following the previous editions of the shared task, systems will be evaluated under multiple terminology modes to assess the causal effect of the explicit terminology contents:
* No terminology * Proper terminology * Random terminology Submissions will be assessed using two complementary groups of metrics:
* Overall translation quality, including automatic MT metrics (e.g., BLEU) together with evaluation of grammatical term usage. * Terminology-oriented metrics, including terminology success rate and terminology consistency.
SUBMISSION GUIDELINES Before submitting, we kindly ask participants to:
1. (Recommended) Pre-register using our Google Form to receive reminders and help us estimate participation. 2. Validate system outputs using the official validation script. 3. Submit translations through the official submission form together with a brief system description. Participants are also encouraged to submit a WMT system description paper following the main conference guidelines.
* The participants are allowed to submit systems for one track and/or only subset of the language pairs (however, we encourage you to test your strengths across all language pairs) * Detailed task descriptions, data formats, input/output specifications, and submission instructions are available on the shared task website.
ORGANIZERS Nathaniel Berger, Adrian Charkiewicz, Pinzhen Chen, Thierry Etchegoyhen, Harritxu Gete Ugarte, Kamil Guttmann, Xu Huang, David Ponce, Artur Nowakowski, Fr��d��ric Odermatt, Arturo Oncevay, Kirill Semenov, Dawei Zhu, and Vil��m Zouhar. Contact: Kirill Semenov �� firstname.lastname [at] uzh [dot] ch
WEBSITE https://www2.statmt.org/wmt26/terminology.html We look forward to your participation!